Data.* collections maintenance

[Daan Leijen]

First of all, thank you to everyone who has volunteered to help/maintain
the Data.Set/Map/etc code. I believe that standard and *stable* libraries
for basic data structures are very important to the community, and encourage
any improvements or changes to the code. I agree that the whole show/read
part of the library is not well thought out -- I guess there should be
the standard read/show instances together with some pretty show functions
for debugging/education.

However, keep in mind that the goal and design of the Data.Set/Map/etc
structures were to be a stable and complete concrete design of data
structures with documented worst-case time complexity. It was not designed
as a general data structure interface that could accommodate many different
implementations. i.e.

> What exactly does prohibit the use of a class here?
> What about the class design of Edison (I admit I never used it).
> http://www.haskell.org/ghc/docs/edison/users004.html#toc3

That is why I think that we should not use classes or anything fancy here.
The design of a proper general data structure interface is desirable and a very
interesting challenge, but it should belong in a different module/library
(Data.Generic.Set, Data.Edison, ?) that perhaps uses Data.Set/Map/etc,
or Adrian's AVL modules to provide implementations.

All the best,
-- Daan.

ps. Just to be clear, the code does not belong to me any more and the
libraries are free to use and modify, and I wouldn't mind at all
if someone wants to take a different direction with it (especially
since I can not contribute myself anyway :-).

In general, I think it should be easier for everyone to contribute
code to the library base. This was one of the discussion topics at
the Haskell workshop and I think this is a good case where it seems
to be difficult to contribute easily: there seemed to be many people
that had their own read instances but weren't able to add this code
to the standard libraries -- I am all for anarchy and fun ;-) and I think
the libraries would be better because of it. Maybe we should think about
how patches could be easily contributed -- should we move over to darcs
for the standard library code? Together with clear instructions on how
to send in patches? -- maybe even allow everyone to just commit...




> 
> IMHO the use of interfaces (instead of relying on the discipline
> of library writers and users) is the important step forward
> from the collections framework of STL (C++) to that of Java.
> 
> We certainly don't want to fall behind that,
> or if we do, we need to have very good reasons.